Objective and subjective relationships of non-linguistic sounds
Thomas F Burns · 2019
Our hearing is used to understand speech as well as non-linguistic sounds (NLSs; car horn, laughter, etc.), but how we perceive them and how we correctly identify them appears to differ from the processes used in speech perception. We examined how differences in objective features of different NLSs influence their subjective perception and correct identification.To do this we utilized a database of 158 NLSs which had been characterised by 12 normal-hearing young adults for the subjective perceptions of complexity, pleasantness, and familiarity, and for accuracy of identification. Using various mathematical, acoustical, and statistical approaches from studies of complex waveforms, we extracted 15 objective features from the NLSs. These objective features were then used to correlate with the subjective perceptions and the accuracy of naming for those NLSs. We additionally used a series of multivariate techniques to establish whether combinations of objective features for NLSs could also account for their differences in subjective perceptions or accuracy of naming. Finally, we also noted the categories of NLSs by their source and used ANOVAs to find if significant differences could be found between these categories for objective features, subjective perceptions, or accuracy of naming.Significant (p<0.05) inter-relationships between the perceptions and accuracy of naming were indentified, which allowed us to determine familiarity as the least independent subjective perception in our study, i.e., it was highly correlated with the other perceptions and accuracy of naming. Significant (p<0.05) inter-relationships between several objective measures were also found, allowing us to identify the salient objective measures. We found several salient objective measures which significantly (p<0.05) correlated (individually) with the perceptions and accuracy of naming for NLSs in our database. Further, combinations of these measures correlated even more significantly; however there were limits to the amount of variation which could be explained with our set of objective measures. Nevertheless, given the objective information of the NLSs, our database could be mathematically categorised into clusters of objectively similar NLSs. These resultant clusters had significant (p<0.05) perceptual differences for complexity and significant (p<0.05) differences in their accuracy of naming. Subjective and objective differences were also noted among different source-based categories of NLSs in the entire database, though there were fewer significant (p<0.05) differences among these categories than there was potential for.Our study marks one of the first and most comprehensive efforts to relate objective features to the subjective perceptions of complexity and familiarity in NLSs, as well as adding substantially to our understanding of the same relationships for pleasantness and for identification accuracy. These findings have implications for the development of novel clinical audiometric testing by providing a well-characterised database of NLSs to test a patient's ability to process spectral and temporal components of sound.